Reproduce “Unified Context Evolution for LLM Agents” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zixuan Zhu, Yitong Hu, Yong Dai, Junfeng Fang, Chunyang Jiang, Senkang Hu, Yuzhi Zhao

The authors propose UCE, an agent that stores reusable notes outside model weights, retrieves relevant notes during tasks, and evolves its library from experience. The notes represent four knowledge types, while a scheduler allocates their generation. In tests on ALFWorld and WebShop, performance improves over ReAct, including 96.3% ALFWorld success versus 75.4%; evidence is limited to two text-based environments.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.